Implementation guide
Automating recruiting workflows without losing human judgment
This guide describes the sequence we use when a company asks us to automate part of hiring. It assumes nothing about tooling and produces a design you can argue with before anything is built.
Published 2026-08-04 · Next review 2027-02-04
1. Write the current workflow down honestly
Not the documented process — the one people actually run, including the spreadsheet, the shared inbox, and the two steps that only work because one person remembers them. Automation applied to an idealised process produces a system nobody uses.
For each step capture the trigger, the input, who does it, how long it waits, and what a failure looks like. Waiting time usually dwarfs handling time, and that alone often redirects the project.
2. Choose the least autonomous design that works
For each candidate step, ask what the simplest mechanism is that would remove the friction. Frequently it is a form, a rule, or an integration rather than a model. Reserve model-based steps for genuinely unstructured work: summarising, drafting, classifying free text.
This is a default we hold across every workflow, not a recruiting-specific caution. Autonomy is a cost, paid in oversight, and it should be bought only where it is earned.
3. Put the gates where the consequences are
Draw the decision points on the map: advance, reject, contact, offer. Each one needs a named owner, a visible record, and a way to reverse it. Everything between the gates can move; the gates themselves stay human.
A useful test: if the system silently did the wrong thing at this step, how long before anyone noticed? If the answer is longer than a day, that step needs a gate or a monitor.
4. Measure cycle time and edit rate, not volume
Volume metrics reward the wrong behaviour — more messages, more screens, more activity. Measure the time from intake to a reviewed shortlist, the proportion of AI-produced artefacts a human materially changed, and the rate of reversals at each gate.
Set the baseline before launch. A workflow with no pre-launch measurement cannot demonstrate improvement afterwards, and the argument for continuing the work becomes a matter of opinion.
5. Name the constraint you actually found
The map often shows that the slow step is not one anybody can automate: interviews that cannot be scheduled because the panel has no availability, or roles that stay open because there is no one to own them. Say so plainly at that point rather than delivering automation around an untouched bottleneck.
Where the constraint is process, this sequence fixes it. Where it is recruiting capacity, that is a different purchase and a different conversation.
Boundary note: Automating a hiring step does not transfer responsibility for it. Employers remain accountable for selection decisions, record keeping, and candidate treatment regardless of which steps a system performs.
Find the AI opportunities worth implementing in your business.
Start with a structured assessment of your workflows, systems, and data, and leave with a prioritized view of where AI can create real value.